A closer look at Agents CLI

Year: 2026 · ▶ Watch on YouTube

Jason Davenport (Technical Lead for Developer Experience) · Shubham Saboo (Senior AI Product Manager)

Switch language → zh

Segments (8)

  • 00:00:00.000 · Introduction — Jason Davenport
    • Jason Davenport introduces the session and his guest, Shubham Saboo, to discuss new AI tools for developers at Google Cloud Next.
  • 00:00:44.500 · Agent CLI: The Entry Point for Agent Development — Shubham Saboo
    • Shubham introduces Agent CLI, a new command-line interface for building, scaling, and governing the entire agent development lifecycle on the Gemini Enterprise Agent Platform.
  • 00:01:50.500 · Understanding Agent ‘Skills’ — Shubham Saboo
    • Explains that ‘skills’ are a standardized way to provide agents with specific capabilities, making them smarter and allowing them to choose which tools to use at runtime.
  • 00:03:38.500 · Simplifying Development with Agent CLI — Shubham Saboo
    • Describes how Agent CLI uses skills to replicate an expert engineer’s workflow, allowing developers to build, deploy, and evaluate agents using simple English prompts.
  • 00:06:35.000 · Agent Development Kit (ADK) 2.0 and Graph-Based Workflows — Shubham Saboo
    • Announces ADK 2.0, which introduces graph-based workflows to bring determinism and reliability to agent processes, crucial for production environments.
  • 00:08:00.500 · Long-Running, Ambient, and Resumable Agents — Shubham Saboo
    • Introduces long-running agents that can maintain state for up to seven days, ambient agents triggered by events, and resumable agents that can recover from interruptions.
  • 00:10:48.000 · Agent Garden: Pre-built Patterns and Templates — Shubham Saboo
    • Presents Agent Garden, a GitHub repository of pre-built agent templates and multi-agent patterns created by Google experts to accelerate development.
  • 00:13:08.000 · Conclusion and Key Takeaways — Shubham Saboo
    • Summarizes the key announcements and emphasizes the shift towards building production-ready, reliable AI agents with a flexible and intelligent developer experience.

Products Announced (4)

  • 00:00:45.000 · Agent CLI (New)
    • Unified CLI for the entire agent development lifecycle (build, scale, govern, optimize). · Works with any coding agent (e.g., Gemini CLI, dot code). · Uses natural language prompts to execute complex agent development tasks.
    • Announced at Google Cloud Next ‘26
  • 00:06:36.500 · Agent Development Kit (ADK) (2.0)
    • Graph-based workflows for deterministic agent behavior. · Support for Ambient Agents (event-triggered). · Support for Resumable Agents (pause/resume functionality).
    • Announced at Google Cloud Next ‘26
  • 00:08:01.000 · Long-running AI Agents (New Capability)
    • Agents can maintain state for up to seven days. · Enables complex, multi-day tasks. · Moves agents from demo-ware to production-ready systems.
    • Available via Agent Runtime, announced at Google Cloud Next ‘26
  • 00:10:48.000 · Agent Garden (New)
    • Library of pre-built agent templates and patterns. · Open source code available on GitHub. · Includes fundamental patterns like sequential, loop, and parallel agents.
    • Announced at Google Cloud Next ‘26

Commitments (1)

  • 00:10:59.000 (In the next month) — Jason Davenport commits to building a golf-swing analysis agent and adding it to Agent Garden.

Notable Quotes (3)

  • 00:00:39.500 — Jason Davenport:

    The power of three, right?

  • 00:12:34.000 — Shubham Saboo:

    We are in the phase when where we are building agents with agents, and all we are doing is talking to agents.

  • 00:13:38.000 — Shubham Saboo:

    We are moving from demo to something that is production, that can run in production reliably and can really solve real problems.

Visual Signals

On-screen (3)

  • 00:00:08.000 · Jason Davenport Technical Lead for Developer Experience, Google Cloud
    • Identifies the host of the segment.
  • 00:00:38.000 · Shubham Saboo Senior AI Product Manager, Google Cloud
    • Identifies the guest speaker and expert on the topic.
  • 00:19:19.500 · Google Cloud Next '26 logo
    • Brands the end of the video segment.

Stage (1)

  • 00:00:00.000 · The two speakers are seated at a desk in a podcast-style booth on the floor of the Google Cloud Next ‘26 conference.

Key Topics

AI Agents · Agent Development · Developer Experience · Google Cloud Next '26 · Agent CLI · Agent Development Kit (ADK) · Gemini Enterprise Agent Platform · Long-running Agents · Stateful AI · Graph-based Workflows · Multi-agent Systems · Agent Patterns · Production AI · Agent Evaluation · Agent Skills

Takeaways

  • Google is launching a suite of tools to move AI agent development from demos to robust, production-ready systems.
  • The new Agent CLI is the central entry point for developers, allowing them to build, govern, and optimize agents across their entire lifecycle using natural language.
  • Agent Development Kit (ADK) 2.0 introduces graph-based workflows, giving developers deterministic control over agent behavior, which is critical for enterprise use cases.
  • A major innovation is long-running agents, which can now maintain state for up to seven days, enabling complex, multi-day tasks and human-in-the-loop processes.
  • ADK 2.0 also adds support for ‘Ambient Agents’ (triggered by events like schedules or texts) and ‘Resumable Agents’ (which can pause and continue after interruptions).
  • Agent Garden provides a library of pre-built templates and multi-agent patterns (e.g., sequential, parallel, human-in-the-loop) to accelerate development and share best practices.
  • The overarching theme is ‘building agents with agents,’ where the development process itself is becoming a conversational experience with AI coding assistants.